ALBERT

Description: Automated bioacoustic analysis aids understanding and protection of both marine and terrestrial \nanimals and their habitats across extensive spatiotemporal scales, and typically involves analyzing \nvast collections of acoustic data. With the advent of deep learning models, classifcation of important \nsignals from these datasets has markedly improved. These models power critical data analyses for \nresearch and decision-making in biodiversity monitoring, animal behaviour studies, and natural \nresource management. However, deep learning models are often data-hungry and require a signifcant \namount of labeled training data to perform well. While sufcient training data is available for certain \ntaxonomic groups (e.g., common bird species), many classes (such as rare and endangered species, \nmany non-bird taxa, and call-type) lack enough data to train a robust model from scratch. This study \ninvestigates the utility of feature embeddings extracted from audio classifcation models to identify \nbioacoustic classes other than the ones these models were originally trained on. We evaluate models \non diverse datasets, including diferent bird calls and dialect types, bat calls, marine mammals calls, \nand amphibians calls. The embeddings extracted from the models trained on bird vocalization data \nconsistently allowed higher quality classifcation than the embeddings trained on general audio \ndatasets. The results of this study indicate that high-quality feature embeddings from large-scale \nacoustic bird classifers can be harnessed for few-shot transfer learning, enabling the learning of new \nclasses from a limited quantity of training data. Our fndings reveal the potential for efcient analyses \nof novel bioacoustic tasks, even in scenarios where available training data is limited to a few samples.